The Accuracy, Completeness, Relevance, Consistency, and Audience-Fit Checklist
CoreEvaluate output accuracy and completeness · Difficulty 2/5
Explanation
A Practical Five-Part Checklist
Once success criteria are set, a practical checklist turns them into a repeatable review:
| Dimension | Question to ask |
|---|---|
| Accuracy | Are the facts, figures, names, and citations correct and verifiable? |
| Completeness | Did it address every part of the request, or quietly drop one? |
| Relevance | Does it answer the actual question, not a nearby one? |
| Consistency | Do the numbers, claims, and recommendations agree with each other and with the source material? |
| Fitness for audience | Is the tone, depth, and format right for who will read it? |
Completeness Is Not the Same as Sounding Complete
A response can read as thorough -- well-organized, confident, detailed -- while quietly skipping one part of a multi-part request. Checking completeness means going back to the original request and confirming each requested part is actually present, not just that the response *feels* comprehensive.
Relevance vs. a Nearby Answer
Claude can produce a fluent, on-topic-sounding answer to a slightly different question than the one asked. Relevance checking means re-reading the actual question and confirming the output answers *that* question, not a plausible neighbor of it.
Common exam traps
- Checking only that the output *sounds* complete instead of confirming each requested part is present against the original request.
- Treating accuracy as the only dimension that matters and skipping consistency, relevance, or audience fit -- all five dimensions can independently fail even when the facts are correct.
Key Takeaways
- The five-part checklist is accuracy, completeness, relevance, consistency, and fitness for audience
- Completeness must be checked against the original request's parts, not against how thorough the answer feels
- Relevance means answering the actual question, not a fluent-sounding nearby one
- Each dimension can fail independently -- accurate facts don't guarantee completeness, relevance, consistency, or audience fit
Related Concepts
Defining Success Criteria Before Evaluating Output
Decide what 'good' looks like before evaluating an output, not after
Fact-Checking and Validation Scaled to Stakes
Validation is a deliberate confirmation step, distinct from improving an output's polish or formatting
Discernment: The AI Fluency Competency Behind This Entire Task Statement
Discernment is the AI Fluency competency for judging output against what was requested, what the sources say, and field standards -- it's the named skill behind the success criteria and five-part checklist taught in this task statement